Mdeberta V3 Base Finetuned Ai4privacy V2

Developed by Isotonic

A PII (Personally Identifiable Information) identification and masking model fine-tuned based on microsoft/mdeberta-v3-base, supporting multilingual privacy data processing

Sequence Labeling

Transformers

Supports Multiple Languages#PII Identification #Multilingual Privacy Protection #High-Precision Entity Recognition

Downloads 380

Release Time : 1/13/2024

Model Overview

This model is specifically designed for identifying and masking personally identifiable information (PII) in text, performing excellently in multilingual environments including English, German, French, and Italian.

Model Features

Multilingual Support

Supports PII identification in multiple languages including English, German, French, and Italian

High-Precision Identification

Achieves 96.83% F1 score and 98.96% accuracy on evaluation datasets

Comprehensive PII Category Coverage

Can identify over 50 different types of personally identifiable information, including names, addresses, phone numbers, financial information, etc.

Flexible Licensing Model

Offers both free and enterprise licensing options to meet the needs of organizations of different sizes

Model Capabilities

PII identification in text

Multilingual text processing

Sensitive information detection

Privacy data masking

Use Cases

Data Privacy Protection

Log Data Masking

Automatically identifies and masks sensitive information in log files

Effectively reduces data breach risks

User Data Anonymization

Processes user-submitted form data by removing or replacing PII

Complies with privacy regulations such as GDPR

Financial Compliance

Financial Transaction Record Processing

Identifies sensitive data such as account information and credit card numbers in transaction records

Meets financial industry compliance requirements

base_model: microsoft/mdeberta-v3-base model-index:

name: mdeberta-v3-base_finetuned_ai4privacy_v2 results: [] datasets:
ai4privacy/pii-masking-200k
Isotonic/pii-masking-200k language:
en
de
fr
it metrics:
accuracy
f1
precision
recall library_name: transformers pipeline_tag: token-classification license: cc-by-nc-4.0

ðŸŒŸ Buying me coffee is a direct way to show support for this project.

mdeberta-v3-base_finetuned_ai4privacy_v2

This model is a fine-tuned version of microsoft/mdeberta-v3-base on the ai4privacy/pii-masking-200k dataset. It achieves the following results on the evaluation set:

Loss: 0.0323
Overall Precision: 0.9636
Overall Recall: 0.9731
Overall F1: 0.9683
Overall Accuracy: 0.9896

Useage

GitHub Implementation: Ai4Privacy

Model description

More information needed

Intended uses & limitations

The license depends on the nature of your business, you're given the right to use Ai4Privacy for your projects. Independent users and small businesses (3 staff maximum) can use Ai4Privacyâ€™s dataset and models in their product and solutions for free, and conditionally for commercial terms. However, larger, profit-oriented organizations require a company license which can be requested by licensing@ai4privacy.com. This two-level system helps fund our project, while keeping the source code accessible and the software free for most users. For the exact usage conditions, please read the detailed terms below.

Free license
Corporate license

Free License

Qualification Criteria

You're qualified to use Ai4Privacy at no cost if you're:

An individual A for-profit entity with a staff count of up to 3 A non-profit or not-for-profit organization

Prohibited Use Cases

It's not permissible to duplicate or alter Ai4Privacy's code with the intention to distribute, sell, rent, license, re-license, or sublicense your own derivative of Ai4Privacy's contribution work.

Warranty Notice

The dataset and code comes "as is", without any express or implied warranties, including but not limited to warranties of saleability, suitability for a specific use, and non-infringement. The authors or copyright holders shall not be responsible for any claim, damage or other liability, whether in a contract, tort or otherwise, arising from or in connection with the dataset and model or its use or other dealings.

Support

Support is offered as best as we can manage through GitHub Issues and our Discord. Corporate licensees receive dedicated support.

Corporate License

Should your entity not fall under the category eligible for a free license, you must procure a corporate license to use Ai4Privacy. This license will authorize you to use Ai4Privacy for the use cases outlined in the free license and provide you with access to priority support..

Please refer to licensing@ai4privacy.com for details on pricing and licensing.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 5e-05
train_batch_size: 32
eval_batch_size: 32
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-06
lr_scheduler_type: cosine_with_restarts
lr_scheduler_warmup_ratio: 0.2
num_epochs: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Overall Precision	Overall Recall	Overall F1	Overall Accuracy	Accountname F1	Accountnumber F1	Age F1	Amount F1	Bic F1	Bitcoinaddress F1	Buildingnumber F1	City F1	Companyname F1	County F1	Creditcardcvv F1	Creditcardissuer F1	Creditcardnumber F1	Currency F1	Currencycode F1	Currencyname F1	Currencysymbol F1	Date F1	Dob F1	Email F1	Ethereumaddress F1	Eyecolor F1	Firstname F1	Gender F1	Height F1	Iban F1	Ip F1	Ipv4 F1	Ipv6 F1	Jobarea F1	Jobtitle F1	Jobtype F1	Lastname F1	Litecoinaddress F1	Mac F1	Maskednumber F1	Middlename F1	Nearbygpscoordinate F1	Ordinaldirection F1	Password F1	Phoneimei F1	Phonenumber F1	Pin F1	Prefix F1	Secondaryaddress F1	Sex F1	Ssn F1	State F1	Street F1	Time F1	Url F1	Useragent F1	Username F1	Vehiclevin F1	Vehiclevrm F1	Zipcode F1
0.0622	1.0	10463	0.0541	0.9247	0.9384	0.9315	0.9770	0.9949	0.9917	0.9812	0.9224	0.9847	0.9592	0.9056	0.9595	0.9802	0.9775	0.9350	0.9971	0.8939	0.7380	0.9664	0.0843	0.9721	0.7784	0.6363	0.9993	0.9877	0.9833	0.9696	0.9866	0.9716	0.9914	0.0	0.8238	0.8025	0.9882	0.9874	0.9878	0.8820	0.9085	0.9869	0.8831	0.8686	0.9984	0.9958	0.9786	0.9971	0.9885	0.9482	0.9455	0.9934	0.9956	0.9860	0.9673	0.9799	0.9916	0.9958	0.9995	0.9875	0.9555	0.9819	0.9311
0.0492	2.0	20926	0.0445	0.9376	0.9494	0.9434	0.9788	0.9970	0.9979	0.9883	0.9492	0.9949	0.9626	0.9548	0.9819	0.9911	0.9922	0.9740	0.9985	0.9057	0.5805	0.9771	0.4872	0.9734	0.8257	0.7479	0.9989	0.9944	0.9960	0.9819	0.9933	0.9958	0.9962	0.1521	0.7969	0.8083	0.9957	0.9972	0.9970	0.9335	0.8953	0.9967	0.8786	0.9232	1.0	0.9980	0.9868	0.9985	0.9967	0.9808	0.9708	0.9969	0.9979	0.9980	0.9921	0.9913	0.9979	1.0	1.0	0.9914	0.9788	0.9859	0.9622
0.0392	3.0	31389	0.0395	0.9458	0.9584	0.9521	0.9815	0.9988	0.9971	0.9861	0.9573	0.9955	0.9764	0.9518	0.9836	0.9914	0.9921	0.9802	0.9985	0.9300	0.8115	0.9862	0.5935	0.9821	0.8684	0.7870	0.9994	0.9972	0.9976	0.9860	0.9956	0.9981	0.9964	0.3002	0.8302	0.7274	0.9863	0.9968	0.9979	0.9528	0.9208	0.9881	0.9195	0.9269	0.9992	0.9980	0.9894	0.9998	0.9984	0.9750	0.9767	0.9971	0.9972	0.9978	0.9927	0.9952	0.9966	1.0	0.9997	0.9962	0.9942	0.9937	0.9653
0.0311	4.0	41852	0.0341	0.9537	0.9667	0.9601	0.9855	0.9998	0.9986	0.9869	0.9545	0.9961	0.9774	0.9632	0.9885	0.9952	0.9940	0.9838	0.9985	0.9392	0.8567	0.9863	0.7015	0.9853	0.9158	0.8687	1.0	0.9979	0.9984	0.9875	0.9971	1.0	0.9962	0.4882	0.8664	0.6966	0.9927	0.9973	0.9967	0.9583	0.9251	0.9979	0.9275	0.9351	1.0	0.9988	0.9932	0.9998	0.9980	0.9852	0.9760	0.9983	0.9980	0.9985	0.9932	0.9952	0.9966	0.9996	1.0	0.9969	0.9955	0.9942	0.9685
0.0196	5.0	52315	0.0323	0.9636	0.9731	0.9683	0.9896	0.9998	0.9973	0.9878	0.9495	0.9932	0.9704	0.9648	0.9887	0.9942	0.9940	0.9820	0.9985	0.9570	0.8750	0.9888	0.7416	0.9819	0.9295	0.8946	0.9998	0.9965	0.9984	0.9886	0.9962	1.0	0.9966	0.6284	0.8884	0.8015	0.9940	0.9973	0.9970	0.9653	0.9109	0.9992	0.9524	0.9347	1.0	0.9984	0.9936	0.9998	0.9992	0.9857	0.9801	0.9988	0.9979	0.9983	0.9944	0.9953	0.9974	1.0	1.0	0.9966	0.9936	0.9917	0.9727

Framework versions

Transformers 4.35.2
Pytorch 2.1.0+cu121
Datasets 2.16.1
Tokenizers 0.15.0

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご